François Cartegnie authored 1 year ago and Steve Lhomme committed 1 year ago

Otherwise ends up setting wrong fourcc combination.
ex: avc1/0 -> H264/0, which is meaning annexb, instead of avc1/0 -> H264/avc1

consistency with es_out_Add

We don't use any other fields from the destination. No need to redo the regions
if other things changed in the source.

On the first call the previous visible dimensions are 0x0.

We never use the output format. We only need to create the regions
once.

The subpicture won't have any region.

We only use the value in SubpictureUpdate() and we know the value
regardless if the src/dst visible dimensions changed or not.

We don't use any other fields from the source. No need to redo the regions
if other things changed in the source.

We don't use any other fields from the source. No need to redo the regions
if other things changed in the source.

We don't use any other fields from the source.

And do use a copy with fields we never use.

The callee may not care at all and we're doing many tests for nothing.

This fixes the following case when resuming from pause:

main audio output warning: playback way too early (-7040339): playing silence
main audio output debug: inserting 337936 zeroes / 7040ms
main audio output warning: playback way too late (7038677): flushing buffers

It was not happening with all aout modules. For example, PulseAudio was
likely updating its timing just after a uncork and just before the core
was playing the next audio sample.

The system time is only used when there is no master in order to get the
reference point (wait_sync_ref). Therefore, it's better to always use
the same vlc_tick_now() to avoid gaps.

In case where the audio is not master and paused (unlikely since audio
is not master in case of live playback, that is generally not pauseable).

Fixes some potential enormous silence:
playback way too early (-9223371872593688789): playing silence

"luckily" the allocation was too big and silently failing, and not causing OOM.

Thomas Guillem authored 1 year ago and Steve Lhomme committed 1 year ago

The core is better to handle pause than a small helper, cf. previous
commit.

Thomas Guillem authored 1 year ago and Steve Lhomme committed 1 year ago

Only flushing the module is not enough, the stream state must correspond to
the flushed state.

It's only used after the definition.

They should be rendered by the pre-renderer (spu_PrerenderText).

The same way it's fixed for other regions.

We don't use it below in spu_SelectSubpictures() because the element is
deleted with the index.

It does the same as spu_channel_Clean():

- spu_PrerenderCancel() on each subpicture
- subpicture_Delete() on each subpicture
- empty the channel->entries vector

Claudio Cambra authored 1 year ago and Steve Lhomme committed 1 year ago

Signed-off-by: Claudio Cambra <developer@claudiocambra.com>